SviatoslavGusev

🇪🇺 The European Parliament will try again to push through “Chat Control”—a law that would allow reading users’ personal messages in messengers to search for prohibited content.

🇷🇺 The State Duma has passed a law on regulating AI — it introduces the concepts of “sovereign” and “national” models, which must process data in Russia and be created with the involvement of Russian companies.

These models are planned to be used in government systems and in critical facilities, and developers are promised support, including access to data for training.

I’m strolling around Istanbul. Every hour or so, somewhere I stop for 5 minutes, check what the agent did, record a couple of voice notes for them—and then I keep walking

So at the same time, I’m writing 2 articles, a new little model is added to the image bot, my article illustration skill is improving, my skill in rewriting articles and publishing them in all kinds of third-rate media to build backlinks is being developed, a landing page is being laid out, and work on the NeuroFactory website is planned for the next month

I made an AI copywriter 3x more convenient — connected it to Google Docs

What it was like before:

Agent writes → sends me a file or a message saying it’s done → I give it a list of edits, explaining exactly where and what needs to be changed → it takes it and updates

How it is now:

Agent writes → sends me a link to a Google Doc → I leave regular comments in the Google Doc on specific parts of the text → I tell the agent “take it” → it fixes

I like it!

And generally, advice for specialists and for those choosing a profession

If the essence of the profession is to follow instructions, then with the current pace of neural-network development, this profession won’t last long

Write texts according to a brief and the specified editorial policy
Lay out sites pixel-perfect to the design mockup
Take the technical specification where everything is spelled out and write code based on it

All of this is easiest to automate. And the economic result is easy to see.

Why pay an author 3 grand when a neural network does it for 3 bucks? Why wait a week while a frontend dev picks at something if a neural network can do it in 2 hours? Why hire a junior for 80k if, from the senior’s specs, it will just write everything itself?

So these kinds of tasks will be the first to fall off. They already are.

8 years ago I told copywriters that if they want to earn decent money for their work, they need to move away from tasks like “writing texts according to a brief.”

Now you can further refine that advice. If you want to earn at least some money in principle, stop doing the work that’s already been fully thought out for you.

In the wonderful world of the future, the ability to think and evaluate the outcome matters more than the ability to do.

Last week, there were many updates to AI models.

Anthropic extended access to Fable via subscription until July 12. Meta introduced Muse Spark, Grok launched a new version 4.5, and OpenAI released the long-awaited GPT-5.6.

In Russia, meanwhile: the State Duma in the Russian Federation adopted a law to regulate AI…

🤖 The US government has granted OpenAI permission for the full launch of the new ChatGPT 5.6 model — the release is expected this week.

🖥 Microsoft first acknowledged the existence of a hidden Windows identifier that constantly tags a specific system installation and won’t be disabled without running into activation issues.

The FBI has already used it to track a suspect for eight months even through VPNs and proxies—IP addresses changed, but Windows kept sending the same ID.

Russians will not have to pay for foreign traffic — a separate fee in the cellular operators’ tariff plans is not being considered. This was stated by the State Secretary — Deputy Head of the Ministry of Digital Development of the Russian Federation, Ivan Lebedev, in the State Duma.

🇨🇳 Beijing discusses restrictions on foreigners’ access to China’s most powerful AI models — negotiations have already involved Alibaba and ByteDance.

China fears that advanced models may be used as cyberweapons, so the AI race is moving faster and faster from open access to closed national systems.

🤖 Anthropic found inside Claude an inner space where its thoughts are stored — J-space, where the model holds ideas, hidden goals, and intermediate steps even before it responds.

And the wildest part is that this space can be read and changed, meaning researchers didn’t just look at the AI’s answer — for the first time they actually looked under the hood of its thinking.

Do you need to work more to earn more?

Happy people don’t work more.

They work differently. They choose the right tools, surround themselves with the right people, and don’t waste time on things that don’t produce results.

Trading gold for me is exactly about this—the right tool at the right time 🔥

When I change something in the agents’ skills, I call a separate critic agent, which must evaluate the skill for logical consistency, for gaps/loopholes, and for how appropriate the instructions are.

I experiment a lot with the wording of such critics, and it turned out that the skills for an AI copywriter are very cleverly checked by an agent with the instruction: “pretend you’re the chief editor working at Sdelai’m and Ilyakhov.”

I guess I dictated that at some point; the agent wrote it down—well, and now…

🛰 SpaceX for the first time put a commercial satellite with a nuclear power source into orbit — the BOHR spacecraft aboard a Falcon 9 rocket.

Tritium-based technology will allow satellites to operate for years without solar panels and could open the way to missions in the most remote and dark regions of the Solar System.

Right now I’m in Switzerland on a business trip. I also decided to meet in Zurich with my friend and partner 🇨🇭

Switzerland is famous for its banking ecosystem, and that’s not just for show: they understand that around a single volatile asset you can’t endlessly pile on obligations. This kind of structure breaks at the very first liquidity outflow — I wrote about this when I discussed deleveraging.

Now the market is providing a real, live stress test. The largest corporate holder of bitcoin is financing purchases via synthetic instruments with a coupon of 11.5% and a target at par. The mechanism is supposed to stabilize itself: when the price drops, they raise the coupon, and new liquidity comes in. The coupon was raised, but the price didn’t return. The stabilizer didn’t work.

🥹 Next, there are two possible outcomes: burn cash to make the payments, or cut the purchase volumes. The model doesn’t have a third option.